Use of Remote Imagery and Object-based Image Methods to Count Plants in an Open-field Container Nursery
Date of Graduation
Master of Science in Horticulture (MS)
Second Committee Member
Third Committee Member
Feature Analyst, MATLAB, Nursery, OBIA, UAV
In general, the nursery industry lacks an automated inventory control system. Object-based image analysis (OBIA) software and aerial images could be used to count plants in nurseries. The objectives of this research were: 1) to evaluate the effect of an unmanned aerial vehicle (UAV) flight altitude and plant canopy separation of container-grown plants on count accuracy using aerial images and 2) to evaluate the effect of plant canopy shape, presence of flowers, and plant status (living and dead) on counting accuracy of container-grown plants using remote sensing images. Images were analyzed using Feature Analyst® (FA) and an algorithm trained using MATLAB®. Total count error, false positives and unidentified plants were recorded from output images using FA; only total count error was reported for the MATLAB algorithm. For objective 1, images were taken at 6, 12 and 22 m above the ground using a UAV. Plants were placed on black fabric and gravel, and spaced as follows: 5 cm between canopy edges, canopy edges touching, and 5 cm of canopy edge overlap. In general, when both methods were considered, total count error was smaller [ranging from -5 (undercount) to 4 (over count)] when plants were fully separated with the exception of images taken at 22 m. FA showed a smaller total count error (-2) than MATLAB (-5) when plants were placed on black fabric than those placed on gravel. For objective 2, the plan was to continue using the UAV, however, due to the unexpected disruption of the GPS-based navigation by heightened solar flare activity in 2013, a boom lift that could provide images on a more reliable basis was used. When images obtained using a boom lift were analyzed using FA there was no difference between variables measured when an algorithm trained with an image displaying regular or irregular plant canopy shape was applied to images displaying both plant canopy shapes even though the canopy shape of `Sea Green' juniper is less compact than `Plumosa Compacta'. There was a significant difference in all variables measured between images of flowering and non-flowering plants, when non-flowering `samples' were used to train the counting algorithm and analyzed with FA. No dead plants were counted as living and vice versa, when data were analyzed using FA. When the algorithm trained in MATLAB was applied, there was no significant difference in total count errors when plant canopy shape and presence of flowers were evaluated. Based on the combined results from these separate experiments, FA and MATLAB algorithms appear to be fairly robust when used to count container-grown plants from images taken at the heights specified.
Leiva, J. N. (2014). Use of Remote Imagery and Object-based Image Methods to Count Plants in an Open-field Container Nursery. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/2108